Stata Accounting ResearchSAFE

🔬 A curated collection of 23,000+ agent skills for empirical research across 8 social science disciplines. | 精选 23,000+ AI Agent 技能库,覆盖8大社会科学学科的实证研究。CoPaper.AI 20分钟完成一篇可复现的规范实证论文,并支持用户上传 Skills。-- Maintained by CoPaper.AI from Stanford REAP.

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: stata-accounting-research
description: "STATA code patterns for empirical accounting and finance research"
metadata:
  openclaw:
    emoji: "📒"
    category: "domains"
    subcategory: "finance"
    keywords: ["STATA", "accounting", "empirical finance", "panel data", "earnings management", "audit"]
    source: "wentor-research-plugins"
---

# STATA Accounting Research Guide

## Overview

Empirical accounting research relies heavily on STATA for data manipulation, statistical analysis, and robustness testing. The field has developed standardized methodological approaches -- earnings quality models, event studies, difference-in-differences for regulatory changes, and instrument variable strategies for endogeneity -- that are implemented in a relatively stable set of STATA patterns.

This guide provides the core STATA code patterns used in top accounting journals (The Accounting Review, Journal of Accounting Research, Journal of Accounting and Economics, and Review of Accounting Studies). These patterns are drawn from commonly used research designs in financial reporting, auditing, tax, and managerial accounting research.

Whether you are estimating discretionary accruals, conducting an event study around an earnings announcement, testing the effect of auditor rotation on audit quality, or implementing a regulatory shock analysis, these patterns provide tested, reviewable STATA implementations.

## Data Preparation

### Loading and Cleaning COMPUSTAT Data

```stata
* ============================================================
* COMPUSTAT Annual Data Preparation for Accounting Research
* Standard preparation used across most empirical accounting papers
* ============================================================

* Load COMPUSTAT annual data
use "compustat_annual.dta", clear

* Keep relevant variables
keep gvkey fyear datadate at sale cogs xsga dp ib oancf act lct che dlc ///
     csho prcc_f ceq re dltt txp xrd ppegt ppent invt rect

* Set panel structure
destring gvkey, replace
xtset gvkey fyear

* --- Basic cleaning ---
* Drop financial firms (SIC 6000-6999) and utilities (SIC 4900-4999)
drop if inrange(sic, 6000, 6999) | inrange(sic, 4900, 4999)

* Require minimum observations
bysort gvkey: gen nobs = _N
drop if nobs < 3
drop nobs

* --- Generate common variables ---
* Total accruals (balance sheet approach)
gen total_accruals = (D.act - D.che) - (D.lct - D.dlc) - dp

* Total accruals (cash flow approach, preferred)
gen total_accruals_cf = ib - oancf

* Scale by lagged total assets
gen lag_at = L.at
gen ta_scaled = total_accruals_cf / lag_at
gen sale_scaled = sale / lag_at
gen ppe_scaled = ppent / lag_at
gen dsale = D.sale / lag_at
gen drec = D.rect / lag_at
gen roa = ib / lag_at

* Market value of equity
gen mve = csho * prcc_f

* Book-to-market ratio
gen btm = ceq / mve

* Leverage
gen leverage = (dlc + dltt) / at

* Firm size
gen size = ln(at)

* --- Winsorize at 1% and 99% ---
foreach var of varlist ta_scaled sale_scaled ppe_scaled roa btm leverage size {
    winsor2 `var', replace cuts(1 99)
}

* Label variables
label var ta_scaled "Total accruals / lagged assets"
label var roa "Return on assets"
label var btm "Book-to-market ratio"
label var leverage "Total debt / total assets"
label var size "Log(total assets)"

save "compustat_clean.dta", replace
```

## Earnings Quality Models

### Modified Jones Model (Dechow et al., 1995)

```stata
* ============================================================
* Modified Jones Model: Estimate discretionary accruals
* Standard model for earnings management research
* ============================================================

use "compustat_clean.dta", clear

* --- Step 1: Estimate non-discretionary accruals by industry-year ---
* Jones (1991) model estimated cross-sectionally

gen inv_lag_at = 1 / lag_at
gen dsale_drec = dsale - drec  // Modified Jones adjustment

* Estimate by 2-digit SIC and year (require >= 15 obs per group)
gen sic2 = floor(sic / 100)

* Cross-sectional estimation
gen da_mj = .
gen nda_mj = .

levelsof fyear, local(years)
foreach y of local years {
    levelsof sic2 if fyear == `y', local(industries)
    foreach ind of local industries {
        * Count observations in this industry-year
        count if sic2 == `ind' & fyear == `y' & !missing(ta_scaled, inv_lag_at, dsale_drec, ppe_scaled)
        if r(N) >= 15 {
            * Estimate Jones model
            quietly reg ta_scaled inv_lag_at dsale_drec ppe_scaled ///
                if sic2 == `ind' & fyear == `y', robust

            * Predict non-discretionary accruals
            quietly predict temp_nda if sic2 == `ind' & fyear == `y', xb
            quietly replace nda_mj = temp_nda if sic2 == `ind' & fyear == `y'
            drop temp_nda
        }
    }
}

* Discretionary accruals = Total accruals - Non-discretionary accruals
replace da_mj = ta_scaled - nda_mj

* Absolute discretionary accruals (common measure of earnings quality)
gen abs_da = abs(da_mj)

label var da_mj "Discretionary accruals (Modified Jones)"
label var abs_da "Absolute discretionary accruals"

save "accruals_data.dta", replace
```

### Performance-Matched Discretionary Accruals (Kothari et al., 2005)

```stata
* ============================================================
* Kothari (2005): Performance-matched discretionary accruals
* Controls for correlation between performance and accruals
* ============================================================

* Add ROA to the Jones model
gen da_kothari = .

levelsof fyear, local(years)
foreach y of local years {
    levelsof sic2 if fyear == `y', local(industries)
    foreach ind of local industries {
        count if sic2 == `ind' & fyear == `y' & !missing(ta_scaled, inv_lag_at, dsale_drec, ppe_scaled, roa)
        if r(N) >= 15 {
            quietly reg ta_scaled inv_lag_at dsale_drec ppe_scaled roa ///
                if sic2 == `ind' & fyear == `y', robust
            quietly predict temp_res if sic2 == `ind' & fyear == `y', residual

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

none-observed

Network

none-observed

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (0)

No findings outside the package's declared scope.

Gates applied: no_behavioural_pass.